JPH0762824B2 - Consultation system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Outline] Regarding a consultation system configured to take in unknown knowledge, a natural language is defined in order to take in new knowledge sequentially and answer new questions. When the natural language interface that converts to an expression and the semantic expression converted by this natural language interface are question expressions, the question expression rules are applied to match, and the knowledge base is searched to display the answer when it is known. , If it is unknown, create a reverse question if necessary, while in the case of learning expression, apply the learning expression rule and search the knowledge base when matching and determine whether the content is unknown Knowledge learning that updates the knowledge base based on the question / learning rules and the results obtained by these question / learning rules and the results of inverse questions Section and an unknown word log section that saves the question sentence in a file when the question contains an unknown word. When the input natural language is a learning expression and a question expression is a reverse question, The knowledge base is updated based on the result, and is configured to be logged when an unknown word is included in the question sentence.

[Industrial application field]

The present invention relates to a consultation system configured to take in unknown knowledge.

[Problems to be solved by conventional technology and invention]

Conventional consultation systems are built to answer questions within the limits of pre-stored knowledge. Therefore, even if research progresses and new research results are obtained,
There was a problem that even if we tried to accumulate this new research result as knowledge on site, we could not do it.

An object of the present invention is to successively take in new knowledge so that new questions can be answered.

[Means for solving problems]

 FIG. 1 shows the principle configuration of the present invention.

In FIG. 1, a display device 1 is used to key in a question sentence or a learning sentence (natural language), and display an answer to a question.

The natural language interface 2 is for converting a natural language such as Japanese mixed with kanji and kana that is key-inputted from the display device 1 into a semantic expression, for example, a list expression.

The question / learning rule 3 applies a question expression rule when the meaning expression converted by the natural language interface 2 is a question expression and matches, and searches the knowledge base 5 to display an answer when known. When it is unknown, a reverse question is created if necessary, while in the case of a learning expression, the learning expression rule is applied and when matching is performed, the knowledge base 5 is searched to determine whether or not the content is unknown. It is a thing.

The knowledge learning unit 4 updates the knowledge base 5 based on the result obtained by the question / learning rule 3 and the result of the reverse question.

The knowledge base 5 accumulates knowledge with a hierarchical frame structure.

The unknown word log unit 6 logs the question sentence in the file 7 when the question contains an unknown term (unknown word).

[Action]

In the present invention, as shown in FIG. 1, a natural language (question sentence or learning sentence) input from the display device 1 is converted into a semantic expression by the natural language interface 2, and the converted semantic expression is converted into a question expression. In this case, matching is performed with the question expression rule, and when the knowledge base 5 is searched and the answer is known, the answer is displayed on the display device 1,
When it is unknown, an inverse question is created as needed and is displayed on the display device 1. Based on the result of the answer to this inverse question, the knowledge base is updated or saved as an unknown word in the file 7 and the administrator later If necessary, the knowledge base 5 is incorporated, and in the case of a learning expression, matching with the learning expression rule is performed, and the knowledge base 5 is searched to update the knowledge base 5 when the content is unknown. ing.

Therefore, in the case of learning expressions, new knowledge is added to the knowledge base 5.
In addition to incorporating the new knowledge into the knowledge base 5 as needed in the case of the answer result to the reverse question and unknown words, the consultation ( Answer questions).

〔Example〕

Next, the configuration and operation of one embodiment of the present invention will be sequentially described in detail with reference to FIGS.

First, by using the question rule example and the operation explanatory diagram of FIG. 3, a natural language question sentence such as “a function for outputting a character string to standard output?” Is keyed in from the display device 1 of FIG. The configuration and operation of the case will be described in detail.

When a natural language question sentence “A function for outputting a character string to standard output?” Shown in FIG. 3 (a) is keyed in from the display device 1, the natural language interface 2 displays the third sentence.
Convert to the list representation (a type of semantic representation) shown in Figure (b). The converted list expression in Fig. 3 (b) has a "?" At the end in the original natural language, so it is determined as a question expression.
Since the original natural language has a "problem", the preceding sentence "output a character string to standard output" is determined to be a function, and the original natural language has a "ha", so the previous sentence It is determined that (noun) is the upper frame. In the case of this example, since it is a question expression, matching is performed with the question rule in the question / learning rule 3 in FIG. 2, for example, the question rule in FIG. 3C, and the question in FIG. It is discriminated as an expression, a knowledge search is performed in FIG. 2, and the result is displayed on the display device 1. Specifically,
The list expression in Fig. 3 (b) is matched with the question rule in Fig. 3 (c), and the 1F sentence part in this question rule matches, so the contents described in the THEN sentence are searched from the knowledge base 5. Then, the function name (printf) in FIG. 3 (d) is obtained. Then, as shown in FIG. 3 (e), the answer is displayed on the display device 1 in the natural language "the function is printif".

If the question contains an unknown term, the second
The answer is NO in the figure, and an inverse question such as "Please tell me how to use printf?" Is displayed on the display device 1 as an inverse question and waits for the instruction from the user. When instructed, it is stored in the knowledge base 5 or stored as the unknown word log 6-1 and later selected by the administrator and loaded into the knowledge base 5.

Second, using the learning rule example and the operation explanatory diagram of FIG. 4, the learning sentence of the natural language from the display device 1 shown in FIG. 2, for example, “the function for outputting the character string to the standard output is prin.
It's tf. A detailed description will be given of the configuration and operation when "" is entered by key input.

When the natural language learning sentence “The function for outputting a character string to standard output is printf.” Shown in FIG. 4 (a) is keyed in from the display device 1, the natural language interface 2 is displayed in FIG. Convert to the list representation shown in (b).
This converted list representation in Fig. 4 (b) is discriminated as a learning representation because the original natural language has "is" at the end,
In the original natural language, the sentence "printf" before "is" is identified as the name, and there is "problem" in the original natural language, so the sentence "print character string on standard output" before this is Since it is determined to be a function and "ha" exists in the original natural language, the sentence (noun) "function" before this is determined to be the upper frame. In the case of this example, since it is a learning expression, the learning rule in the question / learning rule 3 in FIG. 2, for example, the learning rule in FIG. It is discriminated as an expression, and the knowledge is updated in FIG. Specifically, the list representation in Fig. 4 (b) is
Fig. 4 (c) Matched with the learning rule, and the 1F sentence part in this learning rule matches as shown by the dotted arrow in the diagram, so the contents described in the THEN sentence (created contents) As new knowledge, for example, in the knowledge base 5
It is taken in as shown in FIG.

〔The invention's effect〕

As described above, according to the present invention, new knowledge is taken into the knowledge base 5 in the case of the learning expression, and even if it is the question expression, the result of the answer to the reverse question and the case of the unknown word are necessary. Since a structure that incorporates new knowledge into the knowledge base 5 is adopted, it becomes possible to grow as a consultation system (system that answers questions) while successively adding new knowledge during use in the field, and a finite number of knowledge You can build a fully useful system with.

[Brief description of drawings]

FIG. 1 is a block diagram of the principle of the present invention, FIG. 2 is a block diagram of one embodiment of the present invention, FIG. 3 is an example of a question rule and operation explanatory diagram, and FIG. 4 is an example of a learning rule and operation explanatory diagram. In the figure, 1 is a display device, 2 is a natural language interface, 3 is a question / learning rule, 4 is a knowledge learning unit, 5 is a knowledge base, 6 is an unknown word log unit, and 7 is a file.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (56) Bibliography Proceedings of the 33rd IPSJ (Late 1986) P. 1279-1280

Claims (1)

[Claims] 1. A consultation system configured to take in unknown knowledge from the outside, a natural language interface (2) for converting a natural language into a semantic expression, and a meaning converted by the natural language interface (2). When the expression is a question expression, the question expression rule is applied to match, and the knowledge base (5) is searched to display the answer when it is known, and the reverse question when it is unknown,
On the other hand, when the semantic expression is a learning expression, the learning expression rule is applied and matched, and the knowledge base (5) is searched to determine whether the content is unknown / a learning rule (3)
And the knowledge base (5) based on the result obtained by this question / learning rule (3) and the input result for the inverse question.
The knowledge learning unit (4) for updating the question sentence and the unknown word log unit (6) for saving the question sentence in the file (7) when the question sentence contains an unknown word are provided. The knowledge base (5) is used when the result of converting to a semantic expression is a learning expression and its contents are unknown, or when the result of a reverse question to the question expression is the learning expression and its contents are unknown. A consultation system that is configured to be updated and configured to log the question text when the question text contains an unknown word.